PixelRAG

• Source: StarTrail-org/PixelRAG (sent by owner, 2026-08-25)
• Verdict: keep
• Why: Visual RAG done right — renders web pages and PDFs to screenshot tiles, then searches over the images with a LoRA-fine-tuned Qwen3-VL embedding model. Text parsing loses tables and layout; this keeps them. Ships a pre-built index of 8.28M Wikipedia pages you can query with zero setup. Academic work from Berkeley SkyLab/BAIR/NLP group, Apache-2.0, 9.7k stars. Interesting for anyone building search over documents where visual structure matters.